Adaptive Understanding of Big Data for Smart Systems

这项工作将为促进智能信息系统更好的用户体验和更高的性能提出新的方法。智能系统在搜索、推荐、健康、游戏等领域的应用影响着数十亿人的生活。当用户与系统交互时，随着时间的推移，用户的兴趣和系统的内容都涉及到。因此，系统的响应应该适应这些变化。本研究将产生一个新的自适应框架来捕获所有必要的信息，提取知识，然后转换为最优输出。在这个过程中，彻底了解系统中的数据是非常关键的。大数据的动态性来自于系统内大量的项目、用户及其交互。这种大数据对于智能系统的特点包括以下几个方面。一是系统中的项目不定期更新，即大量的新项目同时加入到系统中，现有的项目也被更新或淘汰。其次，对系统的新用户应采取适当的冷启动策略，以获得更好的体验。第三，系统的发展、环境的趋势、现有用户的兴趣都是不断变化的。该项目的预期成果是：1)生成和预测方法的理论进步，包括深度学习和强化学习算法，以理解与智能系统相关的大数据；2)改进建模和数据挖掘工具，加强对具有动态特征的大数据集的分析；3)开发智能系统的框架，该框架包含几个能够分别处理物品、用户和环境的关键组件。同时，这些组件将相互通信，共享发现的知识。所提出的方法将首先以单个条目的形式从现有数据中学习全部信息以及交互动作序列，然后进行调整以产生更好的未来动作。学生们将在这个项目中发展他们的专业知识和处理从真实用户收集的大数据集的能力。知识将在学术会议和期刊上以及与行业利益相关者共享。

This work will propose novel approaches for promoting better user experience and higher performance for smart information systems. Billions of people's lives are affected by smart system applications in the areas including searching, recommendation, health, gaming, etc.. When the user interacts with the system, both the user's interests and the system's contents are involving over the time. Therefore, the system's responses should be adaptive to these changes. This study will produce a novel adaptive framework to capture all the necessary information, extract knowledge, and then convert to the optimal outputs. In this process, it is very crucial to thoroughly understand the data in the system. The dynamic nature of the big data comes from the large number of items, users, and their interactions within the system. The characteristics of such big data for a smart system include the following aspects. First, the items in the system update from time to time that a large number of new items are added to the system simultaneously and the existing items are also updated or eliminated. Second, the new users of the system should be treated with proper cold start strategies for better experience. Third, the development of the system, the trend of the environment, the existing users' interests are all kept changing. The anticipated outcomes of this project are: 1) the advancement of theories of generative and predictive approaches including deep learning and reinforcement learning algorithms to understand the big data associated with the smart systems; 2) improved modelling and data mining tools to strengthen the analysis of large data sets with dynamic characteristics; 3) the development of a framework of the smart system with several key components that are able to treat items, users and the environment separately. Meanwhile, these components will communicate with each other and share the discovered knowledge. The proposed approaches will first learn the full information from the existing data in the form of individual entries as well as sequences of interactive actions, and then adapt to generate better future actions. Students will work on the project developing their expertises and ability to work with big data sets collected from real users. Knowledge will be shared in academic conferences and journals, as well as with the industrial stakeholders.